This is a wiki-page covering all you need to know about running and diagnosing jobs on Coma, the main computing workhorse shared by the cosmologists at the CMU McWilliams Center.
Coma currently has in total 1144 cores, distributed among 61 different nodes. It has several large nodes for shared memory (OpenMP) applications, and allows for communication between nodes for applications using MPI
You can use ssh to log-in to logon machine (a.k.a., the head node)
user@local:~$ ssh -X [email protected]
"The first rule of coma club is to never run jobs on the logon machine!" --- the Dictator
The head node should be used only to write and edit code, compile programs, and copy data between drives on coma or on external machines. It should NOT be used to perform computation-heavy jobs, since this could potentially crash the entire system. Even if it does not crash the machine, it slows down the head node for all users.
-
I am a student/postdoc/faculty member at CMU. How do I set up an account on Coma?
In addition to an Andrew ID, you will need to apply for an SCS Cluster User account. Note that if you are a student or a postdoc you will need to ask your supervisor to sponsor you. That done, you should be able to log in as described above.
-
How do I get started?
See Getting Started for information that you should know before you can use the cluster for the first time.
-
I want to use IDL (or python2.7) but it doesn't seem to be in my
PATH
, should I install them locally?Coma does have both installed system-wide, and many more goodies: read the Getting Started page more carefully and look for the keyword "module".
-
What queues and nodes should I use?
See Queues and Nodes for a table of all the available computational resources on Coma.
-
What is Slurm?
See Scheduling Tips for information about submitting jobs and some useful commands when using the Slurm workload manager.
-
Where should I store my data?
See Storage for information about all the hard drives on Coma.
Please direct help requests regarding this cluster to [email protected].
There is also a slack channel (#coma on astropgh.slack.com), where you can discuss cluster-related issues with fellow Coma users.
contact: [email protected]